Figure 2.

Linear regression evaluation result. It shows the RMSE (a) and the correlation (b) between the prediction values of the linear models (which are based on the cis-regulatory matrices) and the gene expression profiles. The three cis-regulatory element finding approaches are the naive model with simple k-mer counts (denoted by 'cis, simple ct'), our main model with both sequence information and gene expression neighborhood information (denoted by 'cis, coexp'), and the reference model developed by Brohée, et al., 2011 (denoted by 'ref [34]'). The fourth column shows the modeling result for the gene expressions by the original input co-expression network of our method. The legend on the right of each bar chart indicates the number of top k-mers used to build the linear model and predict the expression level (see Methods). A lower RMSE or a higher correlation indicates a better prediction accuracy.

Gao et al. BMC Genomics 2013 14(Suppl 1):S4   doi:10.1186/1471-2164-14-S1-S4